학부생을 위한 강연회: Tipping Point Analysis and Influence Maximization in Social Networks

정교민 (KAIST)
2012. 11. 01.

Diffusion of information, rumors or epidemics via various social networks has been extensively studied for decades. In particular, Kempe, Kleinberg, and Tardos (KDD '03) proposed the general threshold model, a generalization of many mathematical models for diffusion on networks which is based on utility maximization of individuals in game theoretic consideration. Despite its importance, the analysis under the threshold model, however, has concentrated on special cases such as the submodular influence (by Mossel-Roch (STOC '07)), homogeneous thresholds (by Whitney(Phys. Rev. E. '10)), and locally tree-like networks (by Watts(PNAS '02)). We first consider the general threshold model with arbitrary threshold distribution on arbitrary networks. We prove that only if (essentially) all nodes have degrees \omega(log n), the final cascade size is highly concentrated around its mean with high probability for a large class of general threshold models including the linear threshold model, and the Katz-Shapiro pricing model. We also prove that in those cases, somewhat surprisingly, the expectation of the cascade size is asymptotically independent of the network structure if initial adopters are chosen by public advertisements, and provide a formula to compute the cascade size. Our formula allows us to compute when a phase transition for a large spreading (a tipping point) happens. We then provide a novel algorithm for influence maximization that integrates a new message passing based influence ranking and influence estimation methods in the independent cascade model.